Early discernment of breast cancer can significantly improve the prospect of successful recovery and survival, but it takes a lot of time that frequently leads to pathologists disagreeing. Recently, much research has tried to develop the best breast cancer classification models to help pathologists make more precise diagnoses. Consequently, convolutional networks are prominent in biomedical imaging because they discover significant features and automate image processing. Knowing which CNN models are optimal for breast cancer binary classification is crucial. This work proposed architecture for finding the best CNN model. Inception-V3, ResNet-50, VGG-16, VGG- 19, DenseNet-121, DenseNet-169, DenseNet-201, and Xception are analyzed as classifiers in this paper. We have examined these deep learning techniques on the breast ultra-sound image dataset. Due to limited data, a generative adversarial network is used to improve the algorithm’s precision. Several statistical analyses are used to determine the finest convolutional technique for premature breast cancer detection using improved images in binary class scenarios. This binary classification experiment evaluates each strategy across various dimensions to determine what aspects improve success. In both normalized and denormalized conditions, the Xception maintained 95% accuracy. Xception uses the complete knowledge-digging technique and is highly advanced. Therefore, the accuracy is considered to be better than that of others.